Özet:

Signalized intersection management systems have become more efficient with the help of developments in communication and detector systems. Especially, Adaptive Signal Management Systems (ASMS) is planned intersections to adapt to variable traffic parameters. However, previous approaches developed to measure the efficiency of signalized intersections are not compatible with the continuous change of control parameters such as cycle length and phase plans. In this study, a delay estimation model (DÖM) has been developed for intersections managed with ASMS using the deep learning approach. The efficiency of the DÖM has been analyzed by comparing the performances with artificial neural network (ANN) models and analytical models. In addition, DÖM and ANN models were trained with different input variables and their performances were investigated. The data used for modeling were collected from the intersection managed by ASMS in Kırıkkale province by making observations. These observations were made according to vehicle types and delay observations of 6331 vehicles in a total of 487 cycles were made. Analysis results showed that the DÖM model predicts the actual delays with error percentages approximately 2 times lower than the ANN model, and approximately 5 times lower than the analytical models. This study revealed that DÖM is an effective performance measurement model for signalized intersections operating with variable timing.

Anahtar Kelimeler: